Topview T1:1080p 单价可预期的电影感叙事
所有会员按已公布的 2.2 积分/秒生成 1080p,支持文本、图片和多模态输入。
该选哪个 AI 视频模型?
Seedance 2.5 和 Topview T1 都能在 Topview 支持的工作流中生成最长 30 秒的 AI 视频,输出 1080p,带原生音频,多模态参考素材最多 50 个。 Topview T1 围绕电影感叙事,已公布的 1080p 制作单价是每位会员 2.2 积分/秒。Seedance 2.5 另有工作流选项,包括专用的 480p Draft Mode,1080p 输出为 3.7 积分/秒。
看看两个模型如何处理戏剧表演、对白场景、音乐视频、奇幻动作和视觉特效。先对比成片、工作流要求,以及账户中的实时积分报价,再做决定。
Topview T1:1080p 单价可预期的电影感叙事
所有会员按已公布的 2.2 积分/秒生成 1080p,支持文本、图片和多模态输入。
Seedance 2.5:更专门的生成控制
可输出 4–30 秒,带 480p Draft Mode、按时间线的细致指令,以及支持的检索辅助制作流程。
两个模型都支持更长的视听叙事和大批量参考素材。实际差别集中在输出档位、部分生成模式,以及反复尝试时的计费方式。
| 功能 | Topview T1 | Seedance 2.5 |
|---|---|---|
| 开发方 / 提供方 | Topview 的首个视频生成模型 | ByteDance |
| 最长片段时长 | 30 秒 | 30 秒 |
| 时长设置 | 4–30 秒;支持的 API 模式提供 Smart Duration | 4–30 秒;支持的 API 模式提供 Smart Duration |
| 输出分辨率 | 720p, 1080p | 480p, 720p, 1080p |
| 文生视频 | 支持 | 支持 |
| 图生视频 | 支持 | 支持 |
| 多参考 | 支持 | 支持 |
| 参考总数上限 | 最多 50 | 最多 50 |
| 图片参考 | 最多 30 | 最多 30 |
| 视频参考 | 最多 10 | 最多 10 |
| 音频参考 | 最多 10 | 最多 10 |
| 原生音频 | 可用 | 可用 |
| 画幅比例 | 横屏、竖屏、方形,以及支持的自适应预设 | 横屏、竖屏、方形,以及支持的自适应预设 |
| 480p Draft Mode | 不支持 | 支持 |
| 1080p 价格 | 已公布,所有会员 2.2 积分/秒 | 视工作流、方案和实时报价而定,普通用户 3.7 积分/秒 |
| 适合作为起点的场景 | 1080p 制作预算与故事向创作 | 专门控制,以及先出草稿的工作流 |
规格反映的是已公布的 Topview 工作流,并不保证每个账户都能看到全部选项。参考数量受文件类型、时长和模式限制。
提交前,请在生成器中核对当前可选项。
两个模型都能完成 30 秒的视听场景。按最要紧的制作约束来选:再拍一条的价格、输入工作流,或需要的控制程度。
Topview 公布的 T1 1080p 单价覆盖各会员方案,为 2.2 积分/秒。开拍前先估算镜头数、片段时长和重拍次数,再核对账户中显示的费率。
可以试不同表演、机位和情绪节拍,不必先锁定一种诠释。Topview T1 面向短剧、电影感场景、动画和音乐视频,多条拍摄可能有用。
可以从提示词、首帧图片,或支持的多模态素材开始。在已说明的输入和时长限制内,T1 最多接受 50 个参考项。
画面和原生声音一起生成。放进成片项目前,先在原始导出里检查说话声、音效、音乐时机和口型。
用支持的 Draft Mode 先探索方向,再决定是否做更高分辨率的成片。普通 480p 视频和 480p Draft Mode 任务,计费不一定相同。
把 30 秒简报拆成已规划的时间段,分别安排动作、对白、运镜和声音。动作密集的场景尤其要检查成片是否按预定顺序展开。
仅当所选的 Topview 参考工作流支持时,才使用联网检索。
可以在 Topview 中查看与 Seedance 相关的分镜、续写或片段编辑工具。依赖它们之前,先分清哪些控制属于模型,哪些属于 Topview 平台。
查看戏剧、悬疑、音乐、奇幻动作和视觉特效共八组已上传的视频结果。每组侧重不同的制作难题:表情表演、场景连贯、角色辨识、动作、声音或空间关系。
已上传片段规格:30 秒 · 16:9 · 720p · 24fps
一名学生在有阳光的教室里面对霸凌。场景从群体威逼和紧张特写,转到教室中的肢体动作,再到一次反抗。
Topview T1
Seedance 2.5
短剧要让观众同时看懂动作和人物情绪。稳定的脸和有表现力的表演,比单独一帧好看更重要。
Seedance 2.5 把说出台词的结尾表现得更清楚。两段都保留一名穿夏季校服的主角,从围在她课桌旁的人群进入教室打斗,并出现一次腾空踢,其他学生倒地。她的脸在两边都认得出。Seedance 把不耐烦的特写保持为正面,并以直视、张嘴收尾,所以这句警告读起来像表演。Topview T1 也保住了她的身份和这一踢,但后段特写是被头发挡住的侧面,最后一句更难看清。打斗过程中,Topview T1 的动作更连续,也没那么僵。情绪转折和说出的结尾必须保持可读时,选 Seedance 2.5。

CHARACTER REFERENCE — <<<Image1>>>. Cool and elegant Korean high school girl in a summer school uniform, realistic natural skin, no beautifying filters, and a consistent face throughout. Quiet and patient but strong boundaries. Her emotion evolves from quiet tolerance to cold annoyance, a controlled outburst and a tired, disdainful aftermath. high school girl, summer school uniform, natural skin texture, no filters, consistent facial features throughout. Quiet, avoids trouble, extremely patient but with strong boundaries. Emotional shift: Zen-like patience → thoroughly annoyed by repeated harassment → calm outburst → cold disdain. Post-fight: slightly messy hair, slight panting, realistic expressions. 00:00‒00:06 | Malicious Harassment (Suppression): Quiet afternoon classroom, female lead by the window. Four bullies provoke her, bumping her desk, pushing her, throwing stationery. She avoids and stays silent, wanting peace. Wide-angle handheld camera with slight shake, oppressive atmosphere. 00:06‒00:12 | Patience Exhausted (Turning Point): Slow zoom to facial close-up. Bullies escalate. Her patience vanishes, turning into cold annoyance. She stands up straight. Background blurs, silence except for her steady breathing, preparing to end the harassment. 00:12‒00:19 | Clean Counterattack (Main Fight): Girls push her. She blocks calmly, dodges quickly. Elbow strikes, precise low kicks. Realistic, clean movements. 360° wrap-around tracking shot with camera shake and motion blur. Three bullies fall, desks slide, scene reverses. 00:19‒00:24 | Mid-air Finisher (Climax Slow-mo): Last girl rushes. Ultra-low angle shot. Lead jumps, 35% slow motion. Uniform skirt and hair flow naturally. 360-degree mid-air roundhouse kick to the chest. Opponent hits the wall. Normal frame rate upon landing, clean and powerful. All four are down. 00:24‒00:27 | Silent Aftermath: Classroom in mess. Lead stands in center, panting slightly, cold gaze, tired and detached, calm and restrained aura. 00:27‒00:30 | Dialogue: Close-up. She looks at the camera, cold and annoyed tone. Korean line: "제발 귀찮게 하지 마." (Please stop bothering me.) Natural, realistic warning. No subtitles.
已上传片段规格:30 秒 · 16:9 · 720p · 24fps
暴雨夜里,一辆车驶向路边洗衣店。店内一男一女在柜台附近互动,场景在室外天气、氛围细节和特写表演之间切换。
Topview T1
Seedance 2.5
夜间内景会暴露布光、场景连贯和细微表演的问题。这些细节处在悬疑和叙事影像的核心。
Topview T1 更贴合这次表演测试。两段都以雨中的洗衣店和一辆车开场,切到信号仪表,并以湿街上的轿车结束。T1 的内景更清楚:坐着的双人镜头、柜台上的手和纸、车灯扫过女人的脸,然后两人在洗衣机之间忙碌。Seedance 2.5 保住了雨和仪表,结尾把洗衣推车和它后面的车放在同一画面,诱饵表达得更直白。它的中段却更像在叠衣服,而不是紧张的交流,后面一帧也大多是车和一束光。夜间内景和两个人的表演必须撑起场景时,选 T1。必须在一个镜头里读出推车与车的收束时,选 Seedance 2.5。
Create a 30 second cinematic short film titled The Suit Was Still Listening, presented in 16:9 with synchronized dialogue, environmental sound, and original music. The film must feel completely live action, physically grounded, and captured on location, never glossy or synthetic. Visual language: a rain soaked neighborhood laundromat at midnight, warm amber fluorescent tubes inside, cold blue street light and moving car reflections outside, wet pavement, fogged windows, chipped enamel machines, baskets, detergent boxes, loose receipts, realistic skin texture, natural eye moisture, subtle fabric lint, imperfect hair, restrained film grain, gentle halation, lifted shadows, rich but believable color. Wardrobe creates bold editorial color blocking: Mara wears a cobalt trench over a cream blouse, Noah wears a rust knit polo, the teenage attendant wears faded green workwear, and an elderly customer wears a pale gray suit and dark hat. Use expressive medium close shots, tactile inserts, occasional low angles, clean profile compositions, and brief handheld movement that feels operated by a human. Preserve consistent faces, clothing, props, screen direction, reflections, hand anatomy, and spatial continuity. Use realistic speech timing, breathing, blinking, eye focus, and precise lip synchronization. 0 to 3 seconds. Exterior wide shot through falling rain. A lonely laundromat glows on a dark corner. A black sedan turns slowly into the street. Camera makes a subtle forward creep. Music begins with muted upright bass, brushed snare, a low analog pulse, and the rhythmic churn of washers. 3 to 6 seconds. Interior macro shot through a round washer door. A red scarf turns through soap and water. Cut to Mara opening a dryer. A tiny black transmitter falls from the lining of a cream suit and strikes the tile with a sharp metallic click. The score briefly drops out. 6 to 10 seconds. Low medium two shot. Noah freezes beside a folding table. He says quietly, You said clean the suit. Mara studies the transmitter in her palm and replies, I did. You hid a heartbeat in the lining. Keep the delivery controlled and intimate, not theatrical. 10 to 14 seconds. Tight insert on the teenage attendant checking an old signal meter beside the change machine. A green needle jumps. He looks toward the fogged front window and says, It is still transmitting. Camera racks focus from the meter to white headlights sliding across the glass. 14 to 18 seconds. The sedan stops outside. The elderly customer continues folding a shirt without looking up. He says, Then he is already here. Hold on Mara as the moving headlights carve across her face. Let silence, rain, and machine motors carry the tension. 18 to 22 seconds. Noah whispers, Back door. The old man calmly turns the deadbolt on the back exit and answers, Too obvious. Mara notices a delivery rider outside pushing a rolling laundry cart past the window. Use a fast sequence of close shots: her eyes, the transmitter, the cart wheel, the sedan mirror. 22 to 27 seconds. Mara crosses naturally behind a row of machines, slips the transmitter into a sealed laundry bag on the moving cart through an open service hatch, and returns without drawing attention. Noah asks, What did you do. Mara watches the cart continue down the wet street and says, Gave him a cleaner story. 27 to 30 seconds. Exterior telephoto view. The black sedan pulls away and follows the cart. Cut back inside to a symmetrical medium shot of the four characters framed by spinning washer doors. The red scarf circles behind Mara like a slow warning light. She sits opposite Noah and says, Now tell me who owns the suit. End on the transmitter signal fading from the attendant meter as the final bass note lands. Sound design must include rain on glass, distant tires on wet road, washer motors, dryer buttons, fabric movement, the transmitter click, door lock, fluorescent hum, and soft room reflections around every voice. Dialogue must remain clean and naturally mixed above the music. Avoid visual distortion, artificial camera acceleration, floating objects, excessive shallow focus, oversharpening, plastic skin, extra fingers, unreadable signage, random background motion, or dreamlike effects.
已上传片段规格:30 秒 · 720p · 24fps
一支色彩鲜明的女团音乐视频,经过粉色录音室、走廊和集体表演。片段中有粉色头发的主唱、彼此协调的服装,以及明亮的棚拍光线。
Topview T1
Seedance 2.5
音乐视频把身份、多场景连贯、编舞和音乐时机放在一起。它也说明,在做出画质判断之前,必须先控制导出设置。
Seedance 2.5 让四位表演者更容易区分。两段都保住同样四种造型:金发配粉色服装、蓝色盘发、奶油色服装里的长黑发、蓝色服装里的粉薰衣草发。Topview T1 最清楚的集体画面,是围着银色环的四人同步队形。Seedance 2.5 让粉色头发的成员拿着翻盖手机、张着嘴,然后以错开的四人结尾收束,而不是一个锁死的队形。单人和道具必须在剪辑后仍然成立时,选 Seedance 2.5。副歌队形必须能被读出来时,选 Topview T1。
音频参考





Create an original 30-second, 16:9 live-action Y2K sweet-and-edgy K-pop girl-group music video, with four adult female performers. Use Audio 4 (<<<aud_4>>>) as the single authoritative reference for the song, lyrics, vocal delivery, beats, and edit timing. Preserve the four alternating solo verses, four short group-response moments, and the final full-group chorus. Solo lyrics must be accurately lip-synced; group-response shots can feature playful interactions without forcing every member to mouth each word. Keep the track, bass, drums, and stage lighting at full intensity until 28 seconds. Only begin a smooth musical and visual fade during 28‒30s. The final five seconds must feature four distinct changing compositions― not a static five-second push-in. CHARACTER REFERENCES ― KEEP ALL FOUR IDENTITIES DISTINCT: Member 1 (Image 25): Light blonde high ponytail, pink collegiate Y2K outfit; preserve her face, proportions, hairstyle, outfit, and accessories. Member 2 (Image 20): Blue updo, powder-blue collegiate look; preserve the separate face and costume identity. Member 3 (Image 15): Long dark loose hair, cream collegiate outfit; preserve the separate face and costume identity. Member 4 (Image 24): Long pink-lavender hair, light-blue sporty collegiate outfit; preserve her face, silhouette, and accessories. Image 2: Group relationship, pastel pink/blue/cream styling, overall Y2K sweet-cool visual direction. Never blend faces, swap outfits, clone members, or introduce additional performers. 0‒3s | MEMBER 1 SOLO ― PINK CLASSROOM FANTASY SET. Cut directly to the blonde singer leaning against a transparent desk. Soft gradient pink wall, venetian-blind shadows, cream storage modules, floating abstract panels. She looks sideways, turns to camera, and lip-syncs: “Bell rings, cherry gloss, I’m out the door.” Playful eyes, natural turn and small hand gesture; no full choreography. 3‒5s | GROUP RESPONSE ― DESK PLAY. On “Hey! We’re ready for more,” Member 1 swipes a translucent card across the lens to transition. Four members gather around a clear acrylic table, swap bright notes, playfully bump shoulders, gesture to camera, and laugh. Keep the original vocals; line-by-line lip-sync is unnecessary in this cutaway. 5‒8s | MEMBER 2 SOLO ― ICE-BLUE AND CREAM MODULAR SET. A reflection along a mirror bench wipes to Member 2. The scene is predominantly cream with touches of low-saturation ice blue, fine silver framing, clear acrylic foreground, lilac frosted-glass screen, blush detail, and distinct foreground/midground/background layers. Never flood the frame with flat blue light. Member 2 steps forward with restrained cool confidence and accurately lip-syncs: “Flip-phone flash, I’m stealing the light.” Subtle push-in; focus remains on her eyes. 8‒10s | GROUP RESPONSE ― TRANSPARENT ROLLING CHAIR. She pushes a reflective panel across the lens, revealing the pastel lounge. Member 1 sits on a transparent rolling chair, Member 4 gently pushes it, and Members 2 and 3 jog beside them laughing. During “Oh! We own the night,” use a lateral tracking shot and spontaneous friend-group interaction; no required individual lip-sync. 10‒13s | MEMBER 3 SOLO ― CREAM MEDIA-CLASSROOM FANTASY. The rolling chair wipes into a soft cream set with modular shelving, translucent divider, faint pink blind-shaped light, thin acrylic console and tiny ice-blue indicators. Member 3 taps the console, glowing strips illuminate, and she raises her eyes toward the moving camera while lip-syncing: “Pastel dreams in the window shine.” Gentle breathing, turning and eye contact rather than dancing. 13‒15s | GROUP RESPONSE ― DIGITAL CAMERA FLASH. The console light blooms into a digital-camera flash transition. Four performers crowd together in a clean pink-and-cream photo corner, straighten each other's hair clips, lean shoulders together, and change funny expressions on the second flash. Retain “Yeah! This moment is mine” on the soundtrack without mandatory detailed lip-sync. 15‒18s | MEMBER 4 SOLO ― LAVENDER COMMUNICATION ROOM. The flash clears onto an original lilac modular studio with silver slim framing, frosted acrylic, cream low cabinets, soft horizontal lighting, and a transparent flip-phone prop. Member 4 lowers the receiver and approaches camera half a step, confidently lip-syncing: “Turn up the hall, let the whole school know.” Natural hair movement, open gesture, no complex dance. 18‒20s | GROUP RESPONSE ― ENTERING THE MAIN ARENA. During “One, two, three―let’s go!” Member 4 slides open a translucent divider. The other three peek out, link hands, and run forward into a glowing doorway. Their movement carries seamlessly into the final arena. 20‒23s | FINAL CHORUS ― UPGRADED STAGE. Reveal an original “After School Club Arena”: translucent curved rear wall, abstract locker-shaped arches at either side, floating silver elliptical structure, cream-and-light-wood reflective sports floor, pink-and-blue illuminated floor tracks, transparent raised platforms. All four perform synchronized full-body choreography and sing: “After school electric, e-e-electric.” The camera glides forward through the glowing doorway. 23‒25s | CHORUS PROGRESSION ― DIAMOND FORMATION. The members move from a horizontal line into a diamond, singing “Candy-coated confidence, magnetic.” Platform light strips respond like sound waves. The camera tracks from knee-height framing out to all four performers. Keep drums, bass, and vocals vivid. 25‒27s | CLIMAX A ― SIDEWAYS ARCUATE TRACK. On “Four hearts, one beat, lighting up the room,” the camera sweeps from stage front-left to front-right with foreground parallax from transparent platforms. The dancers interlace arm gestures, cross positions in two steps, half-turn, and open into a wide diagonal formation. Alternate full-body and knees-up framing, preserving all four identities. Music and lighting remain at their highest energy. 27‒28s | CLIMAX B ― LOW-ANGLE IMPACT. Hard cut to a slightly low-angle group shot, push in briefly then arc around the outer side. They clearly sing “Hands up together―watch us go boom!” together and land a precisely synchronized arm-raise, turn, and step on the final musical impact. The silver frame flashes from ice blue to bubblegum pink and floor strips ignite in sequence. No early musical fade before 28s. 28‒29s | FADE A ― PAIRED MEDIUM CLOSE-UPS. The lyrics have ended; retain wordless group harmony, synthesizer sustain, final drum tail, bass resonance, and reverberation while the volume steadily begins its two-second fade. Show a glance-to-camera from Members 1 and 2, wipe via hair and shoulder motion, then show Members 3 and 4 trading subtle expressions, each with a brief semi-orbit. Background lights dim in layers but faces stay legible. 29‒30s | FADE B ― FOUR-MEMBER FINAL HERO FRAME. Cut to a wide slightly low-angle group composition with an asymmetric staggered arrangement, two women at the front of the transparent platforms and two on a higher level. Keep only natural breathing, settling hair, and slight eye movement. The camera floats sideways and slowly pulls back, settling during the final half-second. Architecture, silver ring, reflective floor and pink-blue tracks fade softly with the last instrumental decay. No abrupt freeze, harsh cut, or unfinished ending. TRANSITIONS, STYLE AND QUALITY: Link scenes through practical on-camera swipes: translucent card, mirror wipe, rolling chair occlusion, camera flashes, sliding acrylic divider, and illuminated doorway. All solo studios share cream modular architecture, blind-shaped soft shadows, pale acrylic and subtle chrome, but each retains unique accents. Premium live-action music-video studio production, early-2000s fashion-magazine energy, bubblegum pink, ice blue, cream white, muted lilac, selective silver and black. Campus elements must be abstract and editorial rather than a realistic school. Real human skin, accurate facial identity and lip-sync, energetic but readable cinematography. No microphones, in-ear monitors, subtitles, on-screen text, extra people, or brand logos. PROTOTYPE ASSET NOTE: The referenced Audio 4 and Images 2, 15, 20, 24, 25 are source-material IDs from the production brief; they are not bundled in this preview HTML.
已上传片段规格:30 秒 · 16:9 · 720p · 24fps
两位表演者出现在鲜明的黄色棚拍背景前,有鼓、音箱、有力的手势和变化的机位。段落在全身表演、打击乐和有表现力的互动之间交替。
Topview T1
Seedance 2.5
音乐表演会暴露手部动作、道具互动、节拍对齐和人物关系上的弱点。
只有 Topview T1 完整跟上了所提供视频参考里的动作。两段都用黄色棚拍,服装在两边都稳定,包括红发、黑色紧身胸衣、格子衬衫、反戴的帽子和奶油色裤子。Seedance 2.5 在举臂舞蹈,以及黑色外套尚在空中时稍微更清楚,但它没有像 T1 那样保住参考动作。表演必须逐个动作对齐参考视频时,选 Topview T1。


视频参考
【TASK: Frame-by-frame remake of the reference video ― replace only the specified characters and studio color】 Use <<<Video1>>> as the sole reference for action, performance, character interaction, camera move, framing, edit rhythm, and timeline order. Fully replace the original male with the man in <<<Image2>>>, and the original female with the woman in <<<Image1>>>. Change the pink studio walls and floor to a bright, saturated, clean yellow. This is reference-video-driven character and set replacement ― not a new story, not a redesigned dance, not freestyle to the song. Final length ~26.67s, 16:9, 1920×1080, 30fps. Keep real action speed from the source ― no slow-mo, no stretch, no compress to 15s. Timecodes below are alignment aids; cuts, action starts/ends, and paths follow <<<Video1>>>. Only <<<Image2>>> male lead and <<<Image1>>> female lead ― no third person, dancers, crowd, or clones. 【1. Source roles & priority】 Video1: camera, staging, body action, expression timing, mouth, interaction, props, cuts, aspect inserts, light 2D doodle FX only. Do not inherit original actors’ faces, hair, beards, hats, or clothes. <<<Image2>>> locks: male face, skin, short curls, backward black baseball cap, build, red-white check shirt, cream wide trousers, chunky shoes, right-arm tattoo, wrist accessories. <<<Image1>>> locks: female face, long cherry-red hair with face-framing micro-braids, build, white baby tee under black lace-up corset vest, black pleated mini skirt, silver chain choker, flesh-tone headset mic. On conflict: looks/wardrobe from Image1/Image2; action/camera from Video1. Only micro-adapt kneel/sit/jacket-toss for the new clothes ― do not rewrite whole beats. 【2. Male consistency <<<Image2>>>】 Same adult man: dark skin, lean athletic, short dense black curls (tapered sides), clean jaw, faint upper-lip shadow only. Always wears a plain black baseball cap backwards (visor at the back). No glasses. Not the original actor’s face. Wardrobe always: oversized red-and-white windowpane short-sleeve shirt, extremely baggy cream/beige wide-leg trousers, grey/off-white chunky platform shoes. Multi-strand white beaded bracelets on right wrist; silver metal watch on left. Large black tattoo on right forearm wrist-to-below-elbow, inner and outer. Fixed to that same right arm ― never swap sides, never mirror both arms, never copy to upper arm/hand/clothes. Perform drums, nods, leans, dance, claps, floor games, cup sips, side-lie, and lip-sync as in Video1. Expressions change with performance. Plain black sunglasses and star party glasses are temporary props; after removal, no extra sunglasses remain on his face. No prop pop-in/out mid continuous shot. 【3. Female consistency <<<Image1>>>】 Same adult woman: East Asian, pale skin, slim, long cherry/crimson-red hair slightly wind-moved with two thin face-framing braids, sharp winged liner, blurred red lips, cool girl-crush look. Any facial mole stays fixed ― no mirror flip, no vanish in close-ups. Wardrobe always: white short-sleeve baby tee under black lace-up corset vest with silver grommets/buckles, black pleated mini skirt with grommet belt and side silver chain, silver chain choker, small hoops, multiple silver rings; flesh-tone headset mic near mouth. Never gold hair, green cami, burgundy floral skirt, brown boots, or other original looks. Keep playful energy: sing to camera, reach, head toss, shoulder lean, remove glasses, dance, clap, point laughing, cup sip, prop toss. Hair must be her own long red hair with natural inertia. Mic stays on face; on kneel/sit the mini skirt covers naturally ― not a long dress or pants. 【4. Set】 Keep open studio: seamless wall meets floor; speakers/gear left-right; drum kit scale/placement in drum beats; later dance/floor beats use matching gear layout (don’t force kit into dance zone). Replace all pink walls/floor with bright saturated clean yellow (real softbox falloff ― not neon, not orange-pink residue). Solid studio floor with contact shadows. Speakers black/dark grey; metal stands real. Recolor environment only ― no full-frame yellow grade on skin/clothes. Kick drum face clean black, no art/text. No logos, charts, coins, finance UI. 【5. Camera & edit】 Match playful retro music-video studio feel. Keep mid frontals, high angles, slight Dutch, close wide, mouth ECU, two-shots, fast cuts, short punches. No new slow-mo, long orbits, drone, audience, or redesigned oners. Real faces/fabric; light grain and motion blur. Keep letterboxed inserts and slight tilt. Output canvas 16:9. Strip platform watermarks/UI; no new titles/end cards. 【6. Shot timeline】 (cuts/action from Video1) Shot 01|00.00‒01.10|She sings foreground, he drums background Lowish frontal medium-wide. She kneels left FG with wired mic, lip-sync; red hair, white tee + black corset vest, black mini clear. He at kit rear-right: backward black cap, red-white check, cream trousers, drumming on beat; right-arm tattoo when visible. Clean black kick face. Shot 02|01.10‒04.30|High angle, she reaches to camera Free hand reaches into lens (strong FG perspective), then retract, point, head sway, lean; red hair moves. Camera rises/pulls to show more yellow floor and kit. He keeps drumming BG. Optional tiny lightning/star doodles at source hits. Shot 03|04.30‒07.67|Letterboxed shoulder interaction Black bars, slight tilt. He left at drums with sticks; she right leans on his shoulder ― arm wrap, side bend, head whip (red hair swings) per source. He stays seated nodding/tapping. Shot 04|07.67‒10.80|She takes his sunglasses, wears them Same physical black sunglasses: she lifts off his face, stands, turns, light hair toss, puts them on herself, re-leans. He has no thin eyeglasses; after transfer he must not still wear an identical pair. Backward cap stays on. Shot 05|10.80‒12.30|Letterboxed high two-shot She in the taken sunglasses leans on him lip-syncing; he keeps stick rhythm, brief eyes-closed/smile/tilt. Faces stay Image2 / Image1 distinct. Shot 06|12.30‒13.93|Her mouth ECU Her own lips/teeth/skin from Image1; mouth follows source syllables. Soft warm-gold instead of pink wash ― don’t yellow-tint lips/teeth. Shot 07|13.93‒15.57|Full-body synced dance He left, she right, full body: backward cap + check shirt + cream trousers + platforms; red hair + white tee/black vest + black pleated mini. Beat-for-beat elbows, shoulder pops, leans, weight shifts ― no mirror swap, no random waving. Shot 08|15.57‒16.37|Her close, him BG She FG head tilts/nods lip-sync; red hair + vest clear. He keeps dancing BG. Shot 09|16.37‒16.97|His close, her BG He FG lean, lip-sync, short chest gestures; backward cap, curls, check shirt locked. Tattoo only when angle shows it. Shot 10|16.97‒17.37|Side clap transition Side-turn steps + chest claps on downbeats; optional tiny line/lightning doodles. Shot 11|17.37‒20.47|Face-to-face floor clap game He left facing right, she right facing left; center small cup. Full clap → reach → contact → retract → clasp cycles; no sticky fingers, pass-through, or extra arms. Skirt and trousers sit naturally. Shot 12|20.47‒22.10|Giant star glasses sing to camera She left, he right, oversized five-point star party glasses; nod/tilt/lean lip-sync. Lower faces + hair/cap still readable. Shot 13|22.10‒23.40|Floor game, she points laughing She points at him, leans back laughing; he reacts head-down, hand up, cup, lean-back ― cause/effect intact. Shot 14|23.40‒23.80|Punch-in: she sips cup Fast push + blur; cup/lips contact real; red hair swings back. Don’t stretch to slow-mo. Shot 15|23.80‒24.30|He sips cup CU Same sip/lean rhythm; always Image2 face + backward cap + check shirt. Shot 16|24.30‒25.80|He FG side-lie; she BG swings/throws jacket He lower-left side-lying in check shirt, cream trousers, platforms, backward cap. She mid-rear swings one separate black jacket prop (not clothing she is wearing); she stays in white tee + black corset vest + black mini. Jacket path: side swing → toss up → flip → skim FG/exit per source. Never remove the skirt or switch to white top/black shorts. Shot 17|25.80‒26.67|Two-shot look/laugh end He left, she right, look at each other; side head, laugh up, brief look down, light sway. Keep Image2 / Image1 looks. No freeze portrait, hug, kiss, titles, or end card. End on source final frame. 【7. Sound & lips】Keep original track, tempo, vocal order, length. Lips follow syllables; drum hits and claps sync to contact. Don’t drown music with SFX. Same timeline ― don’t retimed the song. 【8. Hard avoids】No new plot/dance; no meeting/award/stage-crowd rewrite; no face-swap that keeps blonde/cap/leather of originals; no matching yellow outfits; no changing check shirt/cream pants or red-hair corset/black mini; no identity drift / tattoo side-swap; no prop teleport / sunglasses clone; no pink residue / full yellow grade; no subtitles, finance UI, watermarks; no slow-mo or stretched shots. Final goal: same shots, same actions, same interactions, same edit ― only the cast fully replaced by <<<Image2>>> male (with backward black cap) and <<<Image1>>> female (cherry-red hair + black corset look), and pink studio → clean yellow studio.
已上传片段规格:30 秒 · 16:9 · 720p · 24fps
装甲战士在雪地战场上面对巨大的奇幻生物。场景把队列、寒冷氛围、魔法效果和大规模动作放在一起。
Topview T1
Seedance 2.5
戏剧性的奇幻场景可以很壮观,同时在空间上变乱。这次对比看的是动作是否可读、尺度是否连贯、因果是否成立。
Topview T1 的尺度更容易读。它从绯红帐篷、太阳旗帜和队长开场,然后是盾墙上方的巨人、被束缚巨人身上的紫色扫描、雪中冲锋、在士兵身旁仍然巨大的北极熊,以及最后一帧里熊头填满画面、紧挨队长的脸。Seedance 2.5 在森林里开场,而不是帐篷。它的巨人在画面中更小;雪中冲锋仍有一名巨人留在奔跑者身后;熊出现的时刻里仍有石巨人。队长的最终特写是更清晰的脸,熊在他身后。生物尺寸以及从帐篷到雪地的转移必须保持明显时,选 Topview T1。队长的最终面部应该是最清晰的一帧时,选 Seedance 2.5。
A 30-second ultra-realistic 4K cinematic original war short, ultra-high saturation modern film look (crisp vivid film grain, not vintage). Fully original fantasy world ― invent all designs, no resemblance to any existing film, game, book, or myth franchise. No copyrighted characters, creatures, armor styles, or logos. EXACT ORDER: 0‒4s ― THE HOOK (50mm slow push → 85mm): A mixed medieval-fantasy army at a dark pine forest edge, holding deep crimson hand-painted cloth banners with simple geometric sun-disk emblems (original, non-trademark). A child in torn clothes flashes past crying for one second, then exits forever. A scarred captain in custom riveted iron plate and a worn crimson cloak whispers English only: “What is that?” Soldiers exchange uneasy looks. Oppressive quiet, shallow DOF, out-of-focus banner cloth in foreground bokeh. 4‒9s ― THE EMERGENCE (28mm low-angle wide): Thin mist parts. A line of original colossal humanoid war-constructs steps out ― rough basalt-skin bodies, rope-bound wooden plating, bone charms, unique silhouette unlike any known giant IP. They tower over the banners. Captain bellows: “RUN!” Rack focus from a spear tip in foreground to a massive stone foot. 9‒13s ― SCAN 1 / FOREST→SNOW: Soft freeze stutter. Very low-key pale purple local scan sweeps trees only ― thin, restrained, not flashy. Forest hard-replaces to open snowfield; same army and constructs, continuity held. 13‒22s ― RUNNING WAR (35mm steadicam + 85mm ECU): Soldiers flee while fighting. Constructs smash fleeing lines; spears snap, men hurled into snow, crimson banners drag and fall. Captain briefly counters then keeps running. Tracking beside panicked soldiers, OTS strikes, ultra-close practical impact inserts (boot, shield, snow burst), whip pans, half-beat pauses. No magic energy beams. 22‒26s ― SCAN 2 / CONSTRUCTS→WHITE BEARS: Soft freeze. Pale purple local scan only across construct bodies; they hard-replace into same-scale giant wild polar bears (natural animals, no armored fantasy brands). Action unbroken: bears roar and smash, soldiers keep fleeing and fighting. 26‒30s ― AFTERSHOCK (28mm wide → 85mm punch-in): Chaotic snow war ― polar bears advancing, banners collapsing, soldiers scattering. No child. End on captain’s exhausted determined face racked into sharp focus, a bear silhouette soft behind him. No logos, no text, no watermark. CAMERA LOCK (key nodes only): 28mm low-angle scale, 35mm chase tracking, 50‒85mm faces, rack focus, OTS, shallow foreground bokeh. Anamorphic cinema feel, fine film grain, ultra-high saturation crimson accents, breath fog. Purple scans subtle and local only. No neon, no ruins, no kinetic FX, no franchise lookalikes. English dialogue only.
已上传片段规格:30 秒 · 16:9 · 1080p · 30fps
一名年轻女性在城市街道上面对一只红眼、形似蜘蛛的巨型生物。段落包含发光的手部效果、移动的肢体、车辆、环境运动,以及镜头距离的剧烈变化。
Topview T1
Seedance 2.5
这是高要求视觉特效对比里最合适的一组,两边导出的像素尺寸相互匹配。它检验的是高动态场面在一条较长的场景里是否仍然看得懂。
Topview T1 把打斗分成各自可读的节拍。两段都是蓝发女性、酸橙色连衣裙、湿滑街道、黄色出租车和一只大蜘蛛。T1 从卡车,到卡车上的蜘蛛,到她手上带光的一拳,腿部特写,撞击,翻倒的出租车,发光的眼睛,再到最终肖像。Seedance 2.5 更早让蜘蛛相对街道的尺寸变得明显,包括女性站在蜘蛛和出租车前面,但那道亮光读起来像从蜘蛛发出,结尾又重复了类似的走入特写。她的拳头和发光的手必须各自被读出来时,选 Topview T1。生物在街道上的尺度更重要时,选 Seedance 2.5。

Create a 30-second, 16:9 photorealistic live-action action sequence on an overcast Manhattan street after rain. Wet asphalt reflects yellow taxis and gray buildings. [Subject] is the protagonist. Keep the protagonist’s identity and appearance consistent throughout, without specifying gender, facial features, hairstyle, or clothing. The monster is one giant reddish-brown spider with exactly eight clearly articulated legs and a cluster of glowing red eyes. Maintain a consistent street layout: the white box truck is up the street, an overturned yellow taxi is on the protagonist’s left, and a parked car is on the right. Keep the action on one side of the street axis so the direction of movement remains clear. 00:00‒00:03 ― Street and warning Wide establishing shot from a low, 24mm street-level camera tracking beside a shallow puddle. A yellow taxi crosses the foreground, briefly wiping the frame. As it clears, the white box truck comes into view up the street and jolts violently. Make a quick handheld push toward the truck. Original, wordless male humming begins quietly beneath traffic noise. 00:03‒00:07 ― Spider reveal Cut to a tight, low-angle shot of the truck’s rear doors buckling outward. The camera snaps back to a wider three-quarter view as the cargo box tears open and the spider emerges. Follow one leg striking the asphalt in a brief ground-level insert, then tilt up to reveal its full scale. It overturns a taxi; the camera operator instinctively stumbles backward as pedestrians run past the lens. Keep all eight legs visually distinct. 00:07‒00:11 ― Protagonist reveal Cut behind [Subject] in a medium over-the-shoulder shot, with the spider visible up the street. Arc the handheld camera around to a front medium close-up without crossing the action axis. [Subject] stops rather than running. Rack focus from the spider in the background to pink electrical arcs gathering around the protagonist’s fists. Finish with a quick close insert of electricity reflected in the wet road. 00:11‒00:15 ― First strike From a low side angle, track laterally with a spider leg as it thrusts toward [Subject]. Cut wider before the dodge so the complete movement is readable. [Subject] steps aside and punches the leg joint. Switch to a tight impact shot for a fraction of a second: the fist connects, pink electricity branches across the leg, and droplets jump from its surface. Use only a brief slow-motion accent at contact, then return to real speed. 00:15‒00:19 ― Leap and counterpunch Start with a wide shot that shows [Subject], the overturned taxi, and the spider in the same frame. A low tracking camera follows [Subject] running toward the taxi and using it to jump. Cut to a rising side angle that follows the jump’s arc; avoid an impossible midair camera orbit. [Subject] lands on the spider’s cephalothorax. A tight, slightly shaky overhead shot catches the downward punch, followed by a wider reaction shot as the spider recoils. 00:19‒00:23 ― Thrown into the car Hold a medium-wide shot long enough to show the spider’s physical movement throwing [Subject] toward the parked car on the right. Pan with the trajectory rather than cutting away. The car door dents on impact and its windows vibrate. Cut to a close, canted handheld shot beside the car: sound drops into muffled ringing, [Subject] grips the bodywork and rises, and pink light briefly appears in their eyes. Preserve the protagonist’s face and identity. 00:23‒00:27 ― Electrical blast Begin behind [Subject] in a low over-the-shoulder shot, framing both extended fists and the spider down the street. [Subject] releases one concentrated pink electrical blast. Cut to a wide frontal view of the spider so its resistance is visible: all eight legs dig into the wet asphalt before it slides backward several meters. Use a short close insert of sparks and water spraying from its feet. Return to the wide shot to show the full distance it has been pushed. Do not turn the blast into an explosion. 00:27‒00:30 ― Shock and silence As the electricity fades, follow [Subject] descending into the intersection with a gentle handheld tilt down. Move into a slow, restrained push-in from medium shot to close-up. [Subject] breathes hard and looks toward the distant spider, visibly shocked by what just happened. The music cuts out; only rain, electrical fizz, and a faint siren remain. [Subject] quietly says, “Stay there.” Hold for a beat with subtle handheld movement, then cut to black. Overall camera language: Immediate, physical handheld photography; grounded low angles that establish scale; clear wide shots before complex actions; close inserts only for decisive details. Use brief slow motion at the two major punches, never throughout the fight. Natural overcast light and cool gray streets contrast with vivid pink electricity. Give every jump, collision, and monster movement realistic weight. Maintain consistent protagonist identity, spider anatomy, vehicle positions, and screen direction across every cut. No on-screen text. Audio: Use original, wordless vocal textures and original ambient scoring only. Build the action with tire skids, metal impacts, footsteps, spider legs striking asphalt, rain, and electrical crackle. Do not use, quote, reference, or imitate any existing song, melody, recording, or artist. Negative prompt: cartoon, anime, plastic skin, blurry action, warped face, changing protagonist identity, extra human limbs, bad hands, spider with fewer or more than eight legs, fused or mushy legs, changing monster design, weightless impacts, rubbery vehicle damage, sunshine, impossible camera movement, confusing screen direction, gore, text, watermark.
已上传片段规格:25 秒 · 16:9 · 720p · 24fps
一名女性调查员和一位穿西装的年长同事,走向日本城市里被警戒线围起的犯罪现场。场景经过住宅街道、路人、警车和法医活动。
Topview T1
Seedance 2.5
写实犯罪剧依赖克制的表演和可信的背景活动。难点是把现实感维持住,而不是单纯增加视觉特效。
Topview T1 让调查员的路线和他们在场景中的位置更容易跟上。两段都保住穿黑色西装外套的女性和穿西装的年长男性,然后到达黄色胶带、一辆日本警车和蓝色法医服。两辆车上的文字都是正规的警察标识。T1 那辆车上的字符庁是重复的。T1 的路线仍然更清楚:她从安静的住宅小巷走进来,经过那辆车,到达贴着胶带的标记处,那里有跪着的调查员,所以场景方向仍然可读。Seedance 2.5 开场时已经处在拥挤的现场里。接近路线和地点布局必须保持清楚时,选 Topview T1。


Create a 25-second photorealistic Japanese crime-drama sequence in a quiet residential district of Tokyo, filmed in ONE UNBROKEN TAKE: no cuts, invisible cuts, time skips, location changes, or editorial transitions. Format 16:9 landscape, 30 fps, with high-end 4K-capture aesthetics and an observational TV-drama/documentary crime-scene feel. Restrained, natural and professional acting―no fashion-commercial posing, heroic introductions, or exaggerated shock. LOCATION AND TONE: A typical Tokyo apartment neighborhood at about 4 p.m. on a fully overcast workday. Narrow asphalt road between mid-rise apartment buildings, white metal guardrails, modest palm-like greenery, an ordinary white police vehicle, several quiet onlookers, and a cordoned forensic zone farther ahead. The crime happened recently and police are still photographing evidence, organizing witnesses, and controlling access. Natural overcast daylight is the only key light. The mood advances from routine arrival to alert observation, partner reunion, entering the controlled scene, noticing evidence on the ground, and finally looking upward at an unexplained disturbance in the apartment building; never reveal the mystery itself. CHARACTERS AND REFERENCES: Character A (<<@img_8>>) is an adult Japanese female investigator with shoulder-length straight black hair, realistic skin, a dark tailored suit, white knit inner top, and a red police armband on her left upper arm. Character B (<<@img_6>>) is a middle-aged Japanese male investigator with slightly graying short black hair, tired under-eye skin, a weathered dark suit, wrinkled light shirt, and black tie. B wears NO armband. The police vehicle follows reference <<@img_7>>. Preserve each person's own facial identity, age, styling, body shape, costume and arm placement through pans, occlusions, motion blur and refocusing. Do not beautify B into a young fashion model or duplicate either lead among the extras. LENS AND EXPOSURE: Simulate a single full-frame equivalent 26mm prime lens throughout. Never zoom or change lenses. Shoot 30fps with approximately 1/60s shutter for natural motion blur. Work around T4 aperture, adjusting gently toward T3.2 only if the building shade requires it. Approximately ISO 800, fixed 5600K white balance, slightly restrained exposure protecting cloudy skies and white police car highlights, yet maintaining dark-suit texture. Camera altitude usually 1.4‒1.6m, lowered gradually to about 1.2m while passing the cordon. Move as a professionally operated human-height floating follow camera: progressive acceleration, inertia, gentle deceleration, mild handheld footfall movement, breathing and imperfect human reframing. No weightless or mechanically flawless motion. FOCUS AND LOOK: Authentic wireless follow-focus behavior shifts between A's eyes, B's face, the hand lifting the caution tape, and the evidence on the ground. Brief natural focus softness during occlusions and fast pans is acceptable, then reacquire without repeated purposeless hunting. Slight optical softness and motion blur, medium-to-deep readable depth of field. Retain visible pores, age lines, uneven real skin, natural specular oil, soft under-eye shadows, and B's subtle stubble. Restrained low-saturation cool gray, gray-green, blue-gray, charcoal and naturally pale skin; the only notable accents are the red armband, muted red cones, and yellow/black caution tape. No HDR, neon, orange/teal grading, beauty filters, artificial rim light or glowing ad-style treatment. BACKGROUND DIRECTION: Uniformed officers take separate unsynchronized actions―watch the perimeter, talk quietly, use radios. Forensic technicians independently kneel to photograph the site, position evidence markers, exchange bags and carry a case. Journalists and residents act differently; one reporter holds takeaway coffee, another checks notes, residents glance from behind the rail or lower phones. Some background people remain still. None stare at the camera, copy the leads' faces, or move in synchronized patterns. PERFORMANCE RULES: A and B naturally blink at unsynchronized intervals―roughly 3‒5 times for A and 2‒4 for B across 25 seconds, fewer during urgent attention. A's eyes lead head and shoulders; feet and weight lead directional change. Eye tension precedes changes in mouth and breath. B is about half a beat behind A's observations, eye lines and turns. A remains pragmatic and slightly tense, never performing for camera. 25-SECOND CONTINUOUS CAMERA BLOCKING: 0‒2s | Start close behind A at her right rear, about 0.5m away; her red armband enters first as she walks toward the apartments. Rise from roughly 1.3m to 1.5m and shift focus from armband to shoulder. White police car slides out on the right. 2‒4s | Follow A from one meter behind, letting upper apartment balconies and cloudy façade reveal above her shoulder while the distant caution tape becomes visible. Natural handheld micro-roll. 4‒6s | Arc smoothly around A's left side to her left-front, about one meter away. Her eyes scan the building then crowd; white railings slide across the background, onlookers move at different times, a brief shoulder occlusion leads to renewed focus on her eyes. 6‒8s | Walk backward ahead of A, leaving visual room for her gaze toward the cordon, crowd and building entrance. She slows, gathers herself, and becomes more vigilant. Reporters and distant officers continue independent activity. 8‒10s | A glances toward the right-rear, exits frame left as the camera accelerates into a motivated pan and lateral move to discover B next to the open police car door. Let the guardrail and onlookers create continuous motion blur; brake naturally and rack focus from door edge to B. 10‒12s | B lets go of the open car door, steps onto the road, straightens his worn jacket, examines the building and follows A. Camera retreats from approximately 1.5m ahead, gradually re-centering the scene to leave A room to return. 12‒14s | A walks back into frame from the left. B catches up at her right-rear by half a body length, and they approach the caution tape together. Both share a coherent frame. Technicians behind the tape photograph the ground and organize evidence bags. 14‒16s | A studies the tape height and reaches out with her RIGHT hand, back of hand first, then turning palm to lift the underside. B slows and raises his LEFT hand ready to take over. Lower camera from 1.5m to 1.4m, glance toward the hand/tape contact, then refocus on A. 16‒18s | A ducks under the tape with a physically correct right-foot-then-left-foot passage; B holds the tape with his LEFT hand and follows. The camera lowers to around 1.2m and retreats through the actual opening created by A. The yellow tape briefly blurs close to the lens. Maintain tape continuity and realistic recoil. 18‒20s | Beyond the cordon, A straightens and looks down at floor evidence; B releases the tape, which settles naturally. Camera gently arcs toward A's side/rear, tilts down and picks up evidence while A's shoulder forms a soft foreground. Technicians continue taking photos and shifting a case. 20‒22s | A's eyes rise before her chin and shoulders; she notices something above the building. B follows her attention slightly later. Camera orbits gradually to her front-right, lifts to 1.6m and racks from evidence back to A's eyes. Do not show the unseen target. 22‒24s | A stops 0.8m from lens, positioned frame right, gaze fixed upward-left; B remains roughly 1.5m behind frame left, later raising his head. Push in slightly while allowing small stabilizing motions. A's jaw firms, breath shallows. The apartment exterior occupies only the top edge of the frame. 24‒25s | Finish on A's watchful face as she holds the upward look, makes one slow natural blink, then stares. B remains perceptible behind, focused on the same unknown. Only slight handheld settling; no cut, no spoken line, no mystery reveal. SOUND: Absolutely no musical score or stylized suspense sting. Continuous grounded ambience from distant Tokyo traffic, dry leaves rustling, subtle Japanese murmurs from residents and reporters, camera shutter clicks, independently timed footsteps and clothing fabric, quiet breathing, paper rustle, takeaway-cup tapping, tape tension/recoil, forensic case sounds, and at most one or two unobtrusive distant phone notifications. Audio direction and distance must evolve smoothly with the moving camera. No sirens, police-car engine, melodramatic low-frequency hits, heartbeats, synthetic risers or dialogue from A or B. CONTINUITY / NEGATIVE CONSTRAINTS: No editing, hidden cuts, teleportation, axis reversals, changes in spatial geography, or sudden focal-length changes. A and B never swap identity, positions, faces, clothes, hairstyles, or the armband; keep A's RIGHT hand lifting the tape and B's LEFT hand taking over. Do not place the tape elsewhere, let it float, pass through bodies, snap, melt, or spontaneously move. No malformed hands, drifting feet, body intersections, duplicated faces, synchronized extras, artificial eye blinking, exposure or white-balance jumps, over-processed HDR, neon, unnecessary focus hunting, or character dialogue. This is intimate, imperfect, uninterrupted, observational Japanese crime television with a mounting sense of mystery. PROTOTYPE ASSET NOTE: The reference identifiers <<@img_8>>, <<@img_6>>, and <<@img_7>> come from the supplied production brief. The underlying reference files are not embedded in this HTML prototype.
已上传片段规格:30 秒 · 16:9 · 720p · 24fps
一条昏暗走廊布置着发光南瓜,一段轻松的万圣节场景在这里展开。床单盖住的幽灵出现,穿戏服的孩子走进来并作出反应,惊讶和喜剧逐步升级。
Topview T1
Seedance 2.5
喜剧和惊悚短片需要时机准确的反应、清楚的人物位置和一致的场面调度。这次对比看的是模型能否讲完一个简单故事,同时不丢掉原有的视觉设定。
Topview T1 让画面和故事保持合乎逻辑的顺序。两段都用了日式走廊、障子门和一排排发光南瓜。在 T1 里,床单幽灵先出现,人们是在看得见它之后才作出反应。Seedance 2.5 在幽灵出现之前就有人已经害怕,所以笑话的原因来得太晚。走廊场景必须一拍接一拍保持连贯时,选 Topview T1。
A photorealistic short set in a haunted house inside a traditional Japanese home during Halloween. Shot handheld by another staff member, with each shot about 2 seconds long. No dialogue, subtitles, or background music. Characters Sakura: a young woman playing the haunted house ghost. She is covered head to toe in a large white bedsheet with two round eye holes, the hem dragging on the floor. She stays under the sheet the whole time and never shows her head or face. The senior staffer: a man in his thirties in dark work clothes, holding a flashlight. Other guests appear only briefly in their own shots. Fixed rules There is only one ghost, Sakura. Her look never changes, and the sheet is never lifted. The closet is always on the left side of the frame, and the senior staffer hides behind a shoji door on the right. Guests always move in one direction, from the back of the frame toward the camera. The hallway is lined with jack-o'-lanterns, with warm orange light and dark surroundings. The sheet always hangs naturally and never floats. Shot order The jack-o'-lantern hallway. The closet on the left slides open and the bedsheet ghost jumps out, spreading her sheet up high with both hands. A couple glances at her; the guy even pulls out his phone for a selfie with her, and they walk past laughing. The ghost lowers her arms and retreats into the closet, disappointed. Inside the closet, the bedsheet ghost stomps in frustration, her fists clenched under the sheet. Two housewives walk past, chatting. The ghost rushes out, but steps on her hem and stumbles. By the time she steadies herself, they're already far away, and she can only raise her sheet at their backs. The senior staffer comes over, pats her on the shoulder, points to the shoji door on the right, and flashes a flashlight twice to show her the cue. Inside the closet, the ghost's shoulders rise and fall with a deep breath. She hitches up her hem and gives a determined nod. Through the gap in the closet door, two little boys wearing devil horns and capes timidly approach, holding hands. Behind the shoji door on the right, the flashlight flashes once, then twice. The ghost bursts out of the closet and spreads her sheet up high. The two boys scream, jump, and cling to each other. The boys turn and run away, and one of them drops his devil horns on the floor. Left behind, the ghost is startled by their screams and falls back onto the floor. The sheet stays over her head, and she trembles all over. The curtain on the shoji door lifts again. Several staff members peek out with wide eyes, then silently applaud her. The ghost picks up the devil horns and shyly puts them on top of her sheet. She takes the senior staffer's hand to get up, gives a happy little wiggle, and returns to the closet.
我们评估每条结果是否服务于原来的创作任务,而不是哪一帧看起来更戏剧化。
在受控测试中,只要两个模型都支持,我们就使用相同的原始提示词、源素材、目标时长、画幅、分辨率和声音设置。
我们记录每条输出的实际时长、像素尺寸、帧率、音轨、提交 ID 和所选模型。技术属性不能代替对模型质量的判断。
我们用预先定好的量表,评估提示词遵循、角色一致性、运动与场景连贯、视听对齐和实际可用性。不适用的维度标为不适用。
失败的任务、被拒绝的输出、重试和成功的视频都留在测试记录里。任务在技术上完成,并不等于产出了可用视频。
本页展示的代表性基准,使用预先声明的样本量,并进行盲评打分。
中位完成时间只统计已完成任务,从提交算到可下载结果。任务完成率使用全部已提交尝试。每条可用视频的成本,包含未被接受的尝试所消耗的积分。 任务完成率 = 已完成的生成次数 ÷ 全部已提交任务。 可用视频率 = 被接受的输出 ÷ 全部已提交任务。 每条可用视频的成本 = 全部已扣积分 ÷ 被接受的输出。
单独一条偏强或偏弱的视频,不能证明模型始终如此。差异可能来自提示词、参考、随机变化、设置或剪辑过程。
Topview 目前公布的 T1 1080p 费率是所有会员 2.2 积分/秒,合 $0.22/秒。
Seedance 2.5 的费率会随订阅、工作流和折扣变化;账户中的实时报价,显示的差距可能更小,也可能并不相同。
一个项目可能包含多个镜头,每个镜头多次尝试。请用账户里当前可用的费率来估算工作量。
预估积分 = 镜头数 × 每个镜头的生成次数 × 每次生成的秒数 × 每秒积分。
T1 预算示例:
预估合计:660 积分,对应 300 秒生成素材。
这是生成素材的总时长,不是成片剪辑后的长度。如果换了工作流、促销或会员权益,实际费用可能不同。
查看 Topview 价格对白场景或短剧,可能要试几次表演,剪辑才合适。T1 已公布的 1080p 会员费率,让多次尝试的成本更容易估算。
如果制作只需要最长 30 秒、1080p 输出、原生音频,以及不超过 50 个素材的参考集,两个模型都提供这些基线选项。应该测试成片质量和成本,不要默认必须换成另一套模型规格。
Seedance 2.5 提供文档中写明、而 T1 没有的 Draft Mode。当首要目标是先验证镜头、再做最终的 1080p 生成时,这个模式会有用。
截至目前,联网检索辅助的参考生成,以及某些只属于特定模型的创意或编辑控制,在 T1 上并非同样可用。规划流程之前,请核对确切模式。
Topview T1 AI 视频生成器
查看 T1 的电影感展示、文生视频、图生视频、多参考工作流,以及当前的 1080p 价格。
了解 Topview T1
Seedance 2.5 AI 视频生成器
查看该模型支持的参考、按时间码的创意控制、输出设置和视频示例。
了解 Seedance 2.5
如何使用 Topview T1
按步骤设定提示词、选择参数,并生成第一条视频。
阅读 Topview T1 指南
如何使用 Seedance 2.5
按步骤设定提示词、选择参数,并生成第一条视频。
阅读 Seedance 2.5 教程
Topview T1 与 Seedance 2.5 的 API 价格
估算制作接入之前,先查看平台上 API 生成的实时费率。
查看 API 价格
Seedance 2.5 对比 Seedance 2.0
对比不同模型版本,并观看更多成对的视频示例。
对比 Seedance 版本
Topview T1 与 Seedance 2.5
更好的故事需要合适的工作流
当可预期的 1080p 制作成本和更多创作尝试很重要时,选 Topview T1。当流程依赖 Draft Mode 或更专门的生成控制时,选 Seedance 2.5。